Language models, which are components of systems such as automated speech recognizers (ASR) assign probabilities to sequences of speech or text, known as n-grams. ASR transcribe utterances into a series of computer readable sounds, which are then compared to a dictionary of words in a given language. The probabilities can be used to help select the most likely transcription of an utterance.